Atomic Matchmaking with Two-Phase Commit
Two-phase commit semantics for match creation that handles player disconnections gracefully.
When to Use This Skill
- Building real-time multiplayer matchmaking
- Need to handle player disconnections during match creation
- Want to avoid orphaned lobbies and stuck players
- Require reliable match notifications
Core Concepts
Matching two players is deceptively hard. Either player can disconnect between being matched and joining. The solution uses two-phase commit:
- Phase 1: Verify both connections are healthy via ping/pong
- Phase 2: Create lobby, send notifications, confirm delivery
- Rollback: On any failure, clean up lobby and re-queue the healthy player
Implementation
Python
from dataclasses import dataclass, field
from datetime import datetime
from typing import Optional, Tuple
from enum import Enum
import asyncio
class MatchStatus(str, Enum):
SUCCESS = "success"
PLAYER1_DISCONNECTED = "player1_disconnected"
PLAYER2_DISCONNECTED = "player2_disconnected"
BOTH_DISCONNECTED = "both_disconnected"
NOTIFICATION_FAILED = "notification_failed"
@dataclass
class MatchTicket:
player_id: str
category: str
queued_at: datetime = field(default_factory=datetime.utcnow)
@dataclass
class MatchResult:
status: MatchStatus
lobby_id: Optional[str] = None
lobby_code: Optional[str] = None
player1_healthy: bool = True
player2_healthy: bool = True
requeued_player: Optional[str] = None
class ConnectionHealthChecker:
"""Verifies player connections are healthy before matching."""
PING_TIMEOUT = 2.0
MAX_ACCEPTABLE_LATENCY = 500 # ms
def __init__(self, connection_manager):
self._manager = connection_manager
async def check_health(self, player_id: str) -> Tuple[bool, Optional[float]]:
if not self._manager.is_user_connected(player_id):
return False, None
success, latency = await self._manager.ping_user(
player_id, timeout=self.PING_TIMEOUT
)
if not success:
return False, None
if latency and latency > self.MAX_ACCEPTABLE_LATENCY:
return False, latency
return True, latency
async def verify_both_healthy(
self, player1_id: str, player2_id: str
) -> Tuple[bool, bool, bool]:
results = await asyncio.gather(
self.check_health(player1_id),
self.check_health(player2_id),
)
health1, _ = results[0]
health2, _ = results[1]
return (health1 and health2), health1, health2
class AtomicMatchCreator:
"""Creates matches with two-phase commit semantics."""
NOTIFICATION_TIMEOUT = 2.0
NOTIFICATION_RETRIES = 3
def __init__(
self,
health_checker: ConnectionHealthChecker,
lobby_service,
queue_manager,
notification_service,
):
self._health_checker = health_checker
self._lobby_service = lobby_service
self._queue_manager = queue_manager
self._notifications = notification_service
async def create_match(
self, player1: MatchTicket, player2: MatchTicket
) -> MatchResult:
# Phase 1: Health Check
both_healthy, health1, health2 = await self._health_checker.verify_both_healthy(
player1.player_id, player2.player_id
)
if not both_healthy:
return await self._handle_health_failure(player1, player2, health1, health2)
# Phase 2: Create Lobby & Notify
lobby = None
try:
lobby = await self._lobby_service.create_lobby(
host_id=player1.player_id,
category=player1.category,
)
await self._lobby_service.add_player(lobby["id"], player2.player_id)
# Notify both players in parallel
notify_results = await asyncio.gather(
self._notify_with_retry(player1.player_id, lobby["code"], player2.player_id),
self._notify_with_retry(player2.player_id, lobby["code"], player1.player_id),
return_exceptions=True,
)
if not all(r is True for r in notify_results):
raise Exception("Notification failed")
return MatchResult(
status=MatchStatus.SUCCESS,
lobby_id=lobby["id"],
lobby_code=lobby["code"],
)
except Exception as e:
# Rollback: delete lobby if created
if lobby:
await self._lobby_service.delete_lobby(lobby["id"])
return await self._handle_phase2_failure(player1, player2, str(e))
async def _handle_health_failure(
self, player1: MatchTicket, player2: MatchTicket,
health1: bool, health2: bool
) -> MatchResult:
if not health1 and not health2:
return MatchResult(
status=MatchStatus.BOTH_DISCONNECTED,
player1_healthy=False,
player2_healthy=False,
)
# Re-queue the healthy player with priority
if health1 and not health2:
await self._queue_manager.requeue_player(player1, priority=True)
return MatchResult(
status=MatchStatus.PLAYER2_DISCONNECTED,
player1_healthy=True,
player2_healthy=False,
requeued_player=player1.player_id,
)
if health2 and not health1:
await self._queue_manager.requeue_player(player2, priority=True)
return MatchResult(
status=MatchStatus.PLAYER1_DISCONNECTED,
player1_healthy=False,
player2_healthy=True,
requeued_player=player2.player_id,
)
return MatchResult(status=MatchStatus.BOTH_DISCONNECTED)
async def _notify_with_retry(
self, player_id: str, lobby_code: str, opponent_id: str
) -> bool:
for attempt in range(self.NOTIFICATION_RETRIES):
try:
success = await asyncio.wait_for(
self._notifications.notify_match_found(player_id, lobby_code, opponent_id),
timeout=self.NOTIFICATION_TIMEOUT,
)
if success:
return True
except asyncio.TimeoutError:
pass
if attempt < self.NOTIFICATION_RETRIES - 1:
await asyncio.sleep(0.1)
return FalseQueue Manager with Priority Re-queue
from collections import deque
from typing import Dict, Optional, Set
import asyncio
class MatchmakingQueue:
"""FIFO queue with priority re-queue support."""
def __init__(self):
self._queues: Dict[str, deque] = {}
self._player_tickets: Dict[str, MatchTicket] = {}
self._lock = asyncio.Lock()
async def enqueue(self, ticket: MatchTicket, priority: bool = False) -> bool:
async with self._lock:
if ticket.player_id in self._player_tickets:
return False
if ticket.category not in self._queues:
self._queues[ticket.category] = deque()
queue = self._queues[ticket.category]
if priority:
queue.appendleft(ticket)
else:
queue.append(ticket)
self._player_tickets[ticket.player_id] = ticket
return True
async def dequeue_pair(self, category: str) -> Optional[tuple]:
async with self._lock:
queue = self._queues.get(category)
if not queue or len(queue) < 2:
return None
ticket1 = queue.popleft()
ticket2 = queue.popleft()
self._player_tickets.pop(ticket1.player_id, None)
self._player_tickets.pop(ticket2.player_id, None)
return ticket1, ticket2
async def requeue_player(self, ticket: MatchTicket, priority: bool = True) -> None:
await self.remove_player(ticket.player_id)
await self.enqueue(ticket, priority=priority)Usage Examples
Match Creation Flow
health_checker = ConnectionHealthChecker(connection_manager)
match_creator = AtomicMatchCreator(
health_checker, lobby_service, queue_manager, notification_service
)
# When two players are matched
result = await match_creator.create_match(ticket1, ticket2)
if result.status == MatchStatus.SUCCESS:
print(f"Match created: {result.lobby_code}")
elif result.requeued_player:
print(f"Re-queued {result.requeued_player}")Best Practices
- Always verify connections before creating lobby - prevents orphaned lobbies
- Re-queue healthy players with priority - they've already waited
- Use notification retries - network can be flaky
- Rollback on any failure - clean up partial state
- Log correlation IDs - essential for debugging match failures
Common Mistakes
- Creating lobby before verifying connections (orphaned lobbies)
- Not re-queuing healthy players (stuck in limbo)
- No notification retries (lost matches on network blips)
- Missing rollback logic (resource leaks)
- Not using priority re-queue (unfair to healthy players)
Related Patterns
- websocket-management - Connection health verification
- distributed-lock - Prevent race conditions in matching
- graceful-shutdown - Drain queue on shutdown